The Accountant We Can't Read Linear A: reading a lost script's arithmetic, not its words

We can voice it. We cannot read it.

Linear A is the writing of Minoan Crete, in use from about 1800 to 1450 BC. Around 1450 BC the Mycenaean Greeks borrowed its signs to write their language; that adaptation is Linear B, and in 1952 Michael Ventris cracked it — because a known language, Greek, sat behind it (the full story is in The Grid That Spoke Greek). Because Linear B grew out of Linear A, scholars lend each Linear A sign the sound of its Linear B look-alike. So we can pronounce Linear A. Here is its most-repeated text, a formula carved on stone offering tables across the island:

That is the whole trick and the whole trap. Reading a sign because it looks like a Linear B sign assumes it sounds like it too — and the sounds may have drifted when the script was borrowed. The transliteration is, in the words of one teaching corpus, “essentially conventional… it is impossible to say whether the homomorphic syllabograms were also homophones.” The result reads aloud and means nothing. The language behind Linear A is unknown. So: read but not understood. Except — for one corner of it.

The numbers need no language

A tally is a tally in any tongue. Linear A counts in a plain decimal system — one mark per power of ten, repeated, and the orders added together. We don't have to know what the scribe called fifty-six to know that it is fifty-six:

Halves and smaller portions get their own signs. Their exact values were a puzzle of their own until 2021, when a Bologna team (Corazza, Ferrara, Montecchi, Tamburini & Valério) recovered them with a computational, statistical and typological model — letting the corpus's own constraints pick the assignment that makes the most distinct fractions expressible. The securest of all, and the only fraction we'll lean on below, is J = ½. The numbers, then, are legible. Which means a column of them can be checked.

The column that checks itself

Most Linear A tablets are accounts: a list of entries, each with a number, and at the bottom one recurring word, KU-RO, followed by a final figure. Add the entries. The sum equals that final figure. Do it on enough tablets and the meaning of the word falls out with no language at all: KU-RO means “total.” Here is the simplest case — six people, each owing a single unit, under the heading KI-RO, “owed”:

Six ones, total six. You just read a word of a dead language off pure addition — the same way the word was first read. As the Linear B co-decipherer John Chadwick put it, “the meaning of one Linear A word is certain: ku-ro… must mean something like ‘total.’” Encyclopaedias call it bluntly “the most secure” word in the script — secured not by grammar but by the fact that “the numeral [after it equals] the sum of the tallies in the list.”

It isn't only people. Here are four consignments of OLIV — olives, a logogram we can read because the same little leafy-twig sign survived into Linear B — summed across a real landholding account, fractions and all:

A note on what's readable here: the numbers and the commodity pictures (olives, wine, figs, grain, a man) are secure — the pictures because they passed into the deciphered Linear B. The words in the left column — RE-ZA, KI-TA-I, SA-RU — are only sounded out by Linear B values and mean nothing to us; they are names of people or places in a lost tongue. We can read the ledger's math and its goods, never its people.

The scribe was off by one

If the sums always landed, you might suspect us of fitting the meaning to the result. They don't always land — and that is the strongest thing on this page. Take HT 13, the famous wine tablet. Add its six entries:

The entries come to 131. The scribe wrote KU-RO 130½. He is short by a half — and the editor of the corpus says exactly that: “KU-RO here records 130.5, but the numbers total 131.” It happens again on HT 122, where the PO-TO-KU-RO — the “grand total” summing two sub-totals — should read 31 + 65 = 96, and instead reads 97:

Think about what catching that requires. You can only find an error in arithmetic you can actually follow. The off-by-one is double proof: proof the scribe was genuinely adding (you can only be “off” against an intended sum), and proof that we are genuinely reading the addition — because a slip is invisible in a system you can't compute. A real human mistake, 3,600 years old, still legible across a language barrier no one has crossed. We are auditing a Bronze Age accountant, and he doesn't always balance.

The same arithmetic gives a second word its meaning, by opposition: where KU-RO is what a list comes to, KI-RO is what is still owed — the deficit. Two words, total and deficit, both deciphered by counting.

Why everything else stays dark

Numbers, commodity-pictures, and a handful of accounting words — KU-RO, KI-RO, PO-TO-KU-RO — are very nearly the whole of what we securely hold. The language is unknown and unclassified, probably a language isolate, and certainly not Greek. Scholars are plain about it:

“the language written in Linear A remains unknown.”Encyclopædia Britannica

“at the moment we can ‘read’ the Linear A script but we still cannot understand it.”Ester Salgarella, Cambridge (creator of the SigLA database)

The proof that it's unread isn't only the silence — it's the noise. A century of confident “decipherments” has read Linear A as half the languages of the ancient world, no two agreeing:

When every key opens a different door, no key fits. The wall has three honest names: no bilingual (no Rosetta Stone), a tiny corpus of short documents (about 1,400, most only a few signs), and no proven relative to lean a guess against. Even the machines hit it. In 1974 David Packard pointed one of the first computers ever turned on an undeciphered script at Linear A — and, tellingly, built a control: nine fictitious decipherments, the Linear B sound-values shuffled at random onto the signs, to test whether the “real” reading was statistically any less meaningless. The modern neural decipherers that cracked Linear B and Ugaritic in 2019 simply don't apply here: they work by leaning on a known relative, and Linear A has none.

That is the exact shape of the boundary. Linear B fell because structure carried you to an anchor and a known language waited behind it. Linear A offers the same structure and the same anchors — the accounts even hand you KU-RO for free — but the language behind the anchor is lost, and structure alone was never enough to raise it. We are left with the strangest of inheritances: a people's bookkeeping, perfectly auditable, their names and their words and their gods sounding in our mouths and meaning nothing. We can read the accountant. We cannot read the man.

The other face of the silence: the libation formula at the top recurs on dozens of altars, and A-SA-SA-RA-ME, its most repeated word, has been guessed to be a goddess's name. Guessed. We don't even know if it names a god.